A few weeks ago, I wrote about an argument I first made in a 2003 Nature essay: that photorealistic scientific images can convey more assuredness than they’ve earned. Pictures rendered with shadows and reflections can look more certain than a Leonardo sketch. But assuredness and accuracy are not the same thing, and today’s far more powerful AI tools make that easier than ever to forget. The software of 2003 now looks almost prehistoric.
Saul Morson and I have just published a piece in The Hedgehog Review that runs the same argument through a different case—Tolstoy, not computer-generated images.
War and Peace is, among other things, an argument against historical determinism. Tolstoy’s generals keep insisting “every contingency has been foreseen,” right before Napoleon destroys them at Austerlitz. The wisest general in the book, Kutuzov, succeeds not because he plans better, but because he notices more. History isn’t a solved equation; it’s generated moment by moment by countless interacting elements, most of whose alternatives vanish without a trace once one path is taken.
Here’s the connection. AI, like the photorealistic image, often projects more assuredness than the underlying reality warrants. It stitches correlations, causal chains, and “foreshadowing” into a single coherent story, and coherence reads as inevitability. The story isn’t false. It’s selective. And the selection is invisible precisely because the result is so fluent.
So, the two pieces share a structure, not a subject: a smooth, confident surface—whether visual or narrative—can project more truth than it contains. The honest move, in both cases, is the one that shows its own provisional nature.
Tolstoy’s antidote isn’t to abandon foresight. It’s to treat it as a compass rather than a map—direction without a predetermined route. As Machado put it, the road is made by walking—not because we lack data, but because the path does not yet exist.
